Semi-Automated Analysis of 4D Flow MRI to Assess the Hemodynamic Impact of Intracranial Atherosclerotic Disease

Abstract

Purpose:

This study evaluated the feasibility of using 4D flow MRI and a semi-automated analysis tool to assess the hemodynamic impact of intracranial atherosclerotic disease (ICAD). The ICAD impact was investigated by evaluating pressure drop (PD) at the atherosclerotic stenosis and changes in cerebral blood flow distribution in patients compared to healthy controls.

Methods:

Dual-venc 4D flow MRI was acquired in 25 healthy volunteers and 16 ICAD patients (ICA, n=3; MCA, n=13) with mild (<50%), moderate (50–69%), or severe (>70%) intracranial stenosis. A semi-automated analysis tool was developed to quantify velocity and flow from 4D flow MRI and to evaluate cerebral blood flow redistribution. PD at stenosis was estimated using the Bernoulli equation. The PD calculation was examined by an in-vitro phantom study against flow simulations.

Results:

Flow analysis in controls indicated symmetry in blood flow rate (FR) and peak velocity (PV) between the brain hemispheres. For patients, PV in the affected hemisphere was significantly (65%) higher than the normal side (P=0.002). However, FR to both hemispheres of the brain was the same. The PD depicted significant correlation with PV asymmetry in patients (ρ=0.67 and P=0.02), and it was significantly higher for severe compared to moderate stenoses (3.73 vs 2.30 mmHg, P=0.02).

Conclusion:

4D flow MRI quantification enables assessment of the hemodynamic impact of ICAD. The significant difference of the PD between patients with severe and moderate stenosis and its correlation with PV asymmetry suggests that PD may be a pertinent hemodynamic biomarker to evaluate ICAD.

Introduction

Intracranial atherosclerotic disease (ICAD) is an etiology for 10–20% of all ischemic strokes (1,2). Despite current aggressive medical therapy (3), approximately 12–22% of symptomatic ICAD patients experience recurrent stroke within the first year of transient ischemic attack or stroke (2,4). Hemodynamic compromise, resulting in low flow and hypoperfusion in distal vascular territories, is considered one of the mechanisms for recurrent stroke in ICAD patients (5).

The Warfarin-Aspirin Symptomatic Intracranial Disease (WASID) trial demonstrated the risk of recurrent stroke to be dependent on the degree of intracranial stenosis (6). However, this methodology did not account for hemodynamic impact or unstable plaque properties of ICAD (7–10), resulting in less precise characterization of disease mechanisms and risk stratification (11,12). To better assess ICAD, local flow rate and velocity across the stenosis and distal vasculature have been evaluated as biomarkers for recurrent stroke (7,13–15). Additionally, it has been shown that intracranial stenosis can globally influence blood flow across the cerebrovascular network (10,16). Pressure drop across the lesion has been proposed for assessing stroke risk (11,12). Patient-specific computational fluid dynamic (CFD) modeling was used to compute the pressure drop across stenosis (17,18). The accuracy of CFD in replicating in-vivo physiological flow depends on computational strategies and modeling assumptions (19). Therefore, a non-invasive diagnostic tool for in-vivo assessment of hemodynamics at the stenosis as well as the cerebrovascular network is optimal for characterizing ICAD.

Blood flow and velocity in stenotic arteries have been measured non-invasively with sonography (13,14,20) or 2D phase-contrast (PC) MRI, also known as quantitative MR angiography (QMRA) (7,15,21,22). QMRA was shown to be effective for flow measurements in intracranial blood vessels (22–24), with low flow states correlating with elevated stroke risk in vertebrobasilar ICAD. However, this technique only offers velocity encoding at fixed spatial 2D locations, making it sensitive to location and orientation of acquisition planes. Although 3D time-of-flight MR angiography (TOF MRA) was employed to facilitate vessel localization and plane placement (22), QMRA is limited by insufficient anatomic coverage, and manual placement of multiple planes is time consuming and unreproducible.

Several studies have demonstrated the feasibility of 4D flow MRI for the evaluation of intracranial hemodynamics (25–29). 4D flow MRI, time-resolved 3D PC MRI with 3-directional velocity encoding, allows velocity measurements across the cerebral vasculature with a single scan (30,31). 4D flow MRI has been previously applied for measuring cerebrovascular flow (27,32–34), and shown to be consistent with 2D PC MRI (35). Pressure drop across the stenosis has been also estimated from 4D flow MRI with the Bernoulli equation or turbulent energy loss (36–38).

The objectives of this study include: 1) development of an automated analysis tool for quantification of 4D flow MRI and 2) applying this analysis tool in a feasibility study to characterize the hemodynamic impact of ICAD. Two parameters are considered: pressure drop (PD) at stenosis and cerebral blood flow redistribution. Global hemodynamic alternations are assessed in patients with ICAD in comparison with controls. Additionally, stenosis PD was estimated in ICAD patients, after examining our method for PD calculation with a dedicated in-vitro flow phantom study against CFD simulations. We hypothesized that cerebral blood flow redistribution and PD at stenoses can be useful in characterizing the impact of lesion in patients with ICAD.

Methods

Study cohort

ICAD patients who received 4D flow MRI as part of clinical ICAD MRI protocol at Northwestern Memorial Hospital during 2017–2018 were retrospectively evaluated in this institutional review board (IRB) approved study. Patients with occluded vessels were excluded, and only patients with stenosis involving larger arteries (diameter > 2 mm, i.e., internal carotid artery, the M1 segment of the middle cerebral artery, or basilar artery) were included.

To establish a baseline for normal cerebral blood flow distribution, 25 healthy volunteers (10 women, 41±18 years of age) with no known history of cerebrovascular disease were included. The volunteer study was approved by the local IRB, and informed consent was obtained from all subjects.

Magnetic Resonance Imaging

Patients underwent clinical ICAD MRI protocol which consisted of necessary clinical sequences. We retrospectively analyzed 3D TOF MRA, post-contrast T1 SPACE (Sampling Perfection with Application-optimized Contrast using different flip angle Evolutions) MRI, and 4D flow MRI. Imaging was acquired with a 3T MR scanner (MAGNETOM Skyra, Siemens, Erlangen, Germany) using a 20-channel head/neck coil with parameters presented in Table 1.

Table 1.

MRI sequence parameters

parameter	3D TOF MRA	3D T1 SPACE MRI	4D flow MRI
TR	21	800	5.7–6.6
TE	3.42	25	3.1–4.4
Flip angle	17°	variable	15°
ETL	-	30	-
venc	-	-	50–60/100–120 cm/s
Voxel size	0.26 × 0.26 × 0.5 mm	0.5 × 0.5 × 0.6 mm	0.8–1.2 mm iso
Matrix size	768 × 624	320 × 352	190–224 × 160–210
slices	209	104	40–44
Temporal resolution	-	-	42–86 ms
Acceleration factor	2	2	2
Scan time	4–5 min	8–10 min	10–15 min

The protocol started with 3D TOF MRA for anatomical imaging and localizing volumetric field of view (FOV) for 3D T1 SPACE MRI and 4D flow MRI. The FOV covered diseased vessels and the Circle of Willis (CoW) including basilar artery (BA) and bilateral intracranial internal carotid arteries (ICA), middle cerebral arteries (MCA), anterior cerebral arteries (ACA), and posterior cerebral arteries (PCA) (Figure 1). The post-contrast T1 SPACE MRI scan was for 3D black blood MR vessel wall imaging (BBMR VWI) which was used for determining the degree of stenosis (39,40). T1 SPACE MRI was acquired ~10 minutes after a single-dose intravenous injection of gadobutrol (1 mmol/ml), a gadolinium-based T1 shortening contrast agent (Gadavist, Bayer HealthCare Pharmaceuticals, NJ, USA). The prospectively ECG-gated k-t GRAPPA accelerated dual-venc 4D flow MRI was acquired post-contrast (~30 minutes) and provided time-resolved MR magnitude and two sets of 3-directional phase images from low and high velocity encoding (venc) (41). High-venc velocity data was to correct the velocity aliasing of fast blood flow velocities in low-venc measurement resulting in a single dual-venc 4D flow MRI data set. Thus, a high-venc that results in no aliasing is desired (41–43). Based on our previous studies on ICAD (16), 120 cm/s for high-venc should satisfy this requirement. The low-venc was set to half of the high-venc to achieve optimal unwrapping while maintaining low noise of low-venc acquisition (41).

Figure 1.

(a) TOF MRI maximum intensity projection showing the volume coverage of the 4D flow MRI and (b) systolic 3D streamlines demonstrating the blood flow field colored by velocity magnitude in an ICAD patient with right ICA stenosis (the location of the lesion is shown by arrows)

Evaluation of the Degree of Stenosis

The degree of stenosis in each patient was evaluated by two experienced neuroradiologists (SAA and RA) from 3D BBMR VWI (44,45) using the WASID trial method (46):

$$\text{degree\hspace{0.17em}of\hspace{0.17em}stenosis}=\left(1-\frac{D_s}{D_p}\right)\times 100\%$$ (1)

where D_s and D_p are diameters of blood vessel at maximal stenosis and proximal to the stenosis, respectively. Mild stenosis was defined as <50%, moderate stenosis as 50–69%, and severe stenosis as ≥70%.

Flow Analysis

4D flow MRI data was corrected for noise and eddy currents (41,47,48), and a 3D PC MR angiogram (PCMRA) was created using pseudo-complex difference algorithm (49) (Figure 2a).

Figure 2.

4D flow MRI quantification including (a) segmentation of PCMRA, (b) segmentation of 3D TOF MRA registered with PCMRA, (c) centerline of major arteries of the CoW, (d) automated placement of multiple 2D analysis planes perpendicular to centerlines, (e) segmentation of lumen on TOF MRA and velocity profile at a 2D plane, (f) schematic 4D flow MRI test result showing median and interquartile range (IQR) of area and flow rate as well as maximum peak velocity of each major artery of the CoW, and (g) schematic diagram of the stenosis showing the location of two planes that were used to quantify velocity and area for pressure drop estimation

Semi-automated Flow Analysis

A novel in-house semi-automated analysis tool was developed in MATLAB (MathWorks, MA, USA) for volumetric quantification of 4D flow MRI using a centerline processing scheme (50). It enables simultaneous evaluation of flow and velocity in the entire cerebral vasculature, and provides a comprehensive report of hemodynamic parameters in the network of cerebral arteries.

The analysis tool allows combining velocity data from 4D flow MRI with anatomical information from 3D TOF MRA (workflow in Figure 2). TOF MRA was registered with 4D flow MRI using rigid registration with FLIRT (FSL, Oxford, UK) (51). Cerebral vasculature was segmented automatically on 3D TOF MRA with a thresholding method (50). Centerline calculation across the entire vasculature was accomplished by skeletonization of segmented 3D vasculature (52). Skeleton points were partitioned to accomplish cerebral arterial branches. Three-dimensional quadratic splines were fitted to center points to obtain centerlines of all arterial branches including BA and bilateral ICA, MCA, ACA, and PCA (Figure 2c).

Multiple equidistant 2D planes were automatically placed perpendicular to each artery’s centerline every 1 mm along each branch (Figure 2d). At every 2D analysis plane following parameters were simultaneously calculated for all arterial branches of the CoW.

1. The vessel wall definition was refined by segmenting the lumen boundary on the registered 3D TOF MRA using thresholding (Otsu method) and watershed (to separate touching vessels) algorithms (Figure 2e). Then, cross-sectional area (mm²) of lumen (region of interest, ROI, for MRI quantification) was calculated at 2D planes as $A={\Sigma }_{i=1}^n{A}_i$ with A_i was the area of voxel i and n total number of voxels within the ROI.

2. To calculate peak velocity (PV) at each plane, first, the systole timepoint was determined as the timepoint when the summation of velocity magnitude in all voxels within the cerebral vasculature is maximum globally. Then, the location of PV at each plane was determined as where the averaged velocity at systole and two adjacent timepoints was the highest within the ROI. Velocity magnitude at that location and at systole timepoint was considered as the PV of the plane.

3. Time-averaged flow rate (FR) in ml/s was calculated within the ROI as $FR=\left(1/T\right){\Sigma }_{j=1}^m{\Sigma }_{i=1}^n\left({\overrightarrow{V}}_{ij}.\overrightarrow{N}\right)A_i$ with ${\overrightarrow{V}}_{ij}$ is velocity vector at pixel i and timepoint j, $\overrightarrow{N}$ is normal vector of the analysis plane, and T is the duration of one cardiac cycle.

Vessel area and FR were calculated as the median area (A) and median FR of each arterial branch. To accomplish vessel PV, first, a median filter (window size of 3) was convolved along the centerline of each branch to remove potential measurement noise in PV (supporting information Figure S1). Then, the vessel PV was calculated as the maximum PV along the arterial branch. The final analysis results were summarized in a schematic intracranial 4D flow MRI report for each subject (Figure 2f).

Manual Flow Quantification

To evaluate the performance of the analysis tool, PV and FR obtained from the semi-automated algorithms were compared with manual measurements from positioning of analysis planes at each branch of the CoW by two observers and using commercial software, EnSight (ANSYS Inc, PA, USA) as described in (16). This comparison was performed in a subgroup of five randomly selected healthy controls.

Flow Distribution in the CoW

The impact of ICAD on cerebral blood flow distribution was assessed by comparing PV and FR between two brain hemispheres for ICAD patients and healthy controls. Two approaches were taken: 1) vessel FR (and vessel PV) ratios, indicating symmetry in various arteries, and 2) hemisphere FR (and hemisphere PV) ratios to assess overall interhemispheric symmetry.

Vessel FR and PV ratios were defined per individual arterial branch by equations (2) and (3).

$$\text{vessel\hspace{0.17em}}PV\text{\hspace{0.17em}ratio}=\frac{PV\left(x\right)\text{on\hspace{0.17em}side\hspace{0.17em}}1}{PV\left(x\right)\text{on\hspace{0.17em}side\hspace{0.17em}}2}\text{\hspace{1em}\hspace{1em}where\hspace{0.17em}}x\in \left\{\text{ICA},\text{\hspace{0.17em}MCA},\text{ACA},\text{\hspace{0.17em}PCA}\right\}$$ (2)

$$\text{vessel\hspace{0.17em}}FR\text{\hspace{0.17em}ratio}=\frac{FR\left(x\right)\text{on\hspace{0.17em}side\hspace{0.17em}}1}{FR\left(x\right)\text{on\hspace{0.17em}side\hspace{0.17em}}2}\text{\hspace{1em}\hspace{1em}where\hspace{0.17em}}x\in \left\{\text{ICA},\text{\hspace{0.17em}MCA},\text{ACA},\text{\hspace{0.17em}PCA}\right\}$$ (3)

where PV(x) and FR(x) means maximum PV and median FR of paired vessel x (ICA, MCA, ACA, or PCA) with bilateral presentation. Side1/side2 are left/right for controls and ipsilateral/contralateral (i.e., affected /nonaffected) for ICAD patients. To mitigate the effects of anatomic variations of the CoW (53), the symmetry ratios were also defined per entire hemisphere.

$$\text{hemisphere\hspace{0.17em}}PV\text{\hspace{0.17em}ratio=}\frac{\max \left(\left\{PV\left(x\right):x\in \left\{\text{ICA},\text{\hspace{0.17em}MCA},\text{\hspace{0.17em}ACA},\text{\hspace{0.17em}PCA}\right\}\right\}\right)\text{on\hspace{0.17em}side1}}{\max \left(\left\{PV\left(x\right):x\in \left\{\text{ICA},\text{\hspace{0.17em}MCA},\text{\hspace{0.17em}ACA},\text{\hspace{0.17em}PCA}\right\}\right\}\right)\text{on\hspace{0.17em}side2}}$$ (4)

$$\text{hemisphere\hspace{0.17em}}FR\text{\hspace{0.17em}ratio=}\frac{\mathrm{sum}\left(\left\{FR\left(x\right):x\in \left\{\text{ICA},\text{\hspace{0.17em}MCA},\text{\hspace{0.17em}ACA},\text{\hspace{0.17em}PCA}\right\}\right\}\right)\text{on\hspace{0.17em}side1}}{\mathrm{sum}\left(\left\{FR\left(x\right):x\in \left\{\text{ICA},\text{\hspace{0.17em}MCA},\text{\hspace{0.17em}ACA},\text{\hspace{0.17em}PCA}\right\}\right\}\right)\text{on\hspace{0.17em}side2}}$$ (5)

where the same convention as of Equations (2) and (3) was applied for sides 1 and 2. Hemisphere PV ratio is the maximum PV within one hemisphere divided by the maximum PV on the contralateral side, and hemisphere FR ratio is total FR to side1 divided by total FR to side2 of the brain.

Pressure Drop Calculation

The regional impact of ICAD was examined by stenosis PD calculated with the Bernoulli equation (54).

$$PD=4V_s^2\left(1-{\left(\frac{A_s}{A_p}\right)}^2\right)$$ (6)

where PD is in mmHg, V_s (m/s) is velocity at the stenosis, A_s and A_p (mm²) are lumen area at maximum stenosis and unconstricted area proximal to stenosis, respectively (Figure 2g). Area (from TOF MRA) at maximal stenosis plane and a plane proximal to stenosis and PV (from 4D flow MRI) at stenosis were evaluated and substituted into Equation (6) to estimate PD.

In-vitro Verification of Pressure Drop calculation

Stenosis Flow Phantom Experiment

An in-vitro flow phantom study was conducted to examine PD from 4D flow MRI against PD computed by CFD simulation for a well-controlled flow in a known geometry. Two phantoms were constructed with an unconstricted diameter of 6.25 mm and 50% and 60% cosine-shaped stenoses (55). Two straight pipes of 6.25 mm diameter and 160 mm length were placed upstream and downstream of the stenosis to ensure fully-developed velocity profile at inlet and to match the geometry of CFD simulation. The phantoms were immersed in a container filled with stationary water, to obtain a static reference for eddy current correction. A steady flow circuit consisting of a gear pump (Burt Process Equipment, CT, USA), a flow meter (OMEGA Engineering, CT, USA), and a real-time PID controller (programmed in LabView, National Instruments, TX, USA) circulated water at a consistent FR. Four different FR of 0.1, 0.2, 0.3, and 0.4 L/min were prescribed to investigate PD at various FRs. 4D flow MRI was acquired using a 3T scanner (MAGNETOM Skyra, Siemens, Erlangen, Germany) and TR/TE = 6.4–7.1/3.4–4.1, FA = 18°, and voxel= 0.8 mm isotropic. Low-venc/high-venc settings were 30/60, 40/80, 50/100, 60/120 for 50% stenosis and 40/80, 60/120, 80/160, and 100/200 for 60% stenosis. To achieve the best results of unwrapping, high-venc was set slightly higher than calculated maximum velocity at the stenosis to ensure aliasing would not occur (42,43). The maximum velocity increased with FR and the degree of stenosis, so the high-venc setting was adjusted accordingly. Area, FR, and PV were quantified with the semi-automated analysis tool and PD was estimated using the Bernoulli equation.

CFD Simulations

Flow in phantoms was also simulated to compute “true” PD. The Reynolds number (Re) was in the range of 340 to 1360, and the transitional Re has been reported as 400 for stenosis with 50% diameter constriction ratio (56). Thus, the Reynolds-averaged Navier-Stokes (RANS) equations with k-ω turbulence model (k and ω are turbulence kinetic energy and specific dissipation rate, respectively) were solved numerically using Fluent 18.2 (ANSYS Inc, PA, USA). Computational mesh consisted of 3 to 4 million anisotropic hexahedral cells refined near the wall, to ensure a dimensionless wall distance less than one, y⁺<1, for accurate modeling of near-wall flow (57). Fully-developed parabolic velocity profiles matched with the FRs for in-vitro experiments (0.1–0.4 L/min) were used as inlet boundary condition. Flow outlet was placed 25 unconstricted diameters downstream from the stenosis with a constant pressure as the boundary condition, and no-slip condition was applied at walls. A dynamic viscosity of 0.001 Pa.s and a density of 1000 kg/m³ were used for water. PD between two points proximal and at the center of stenosis was calculated from CFD and considered as a reference for examining PD from in-vitro scans.

Statistical Analysis

For comparison between manual and semi-automated analysis results as well as inter-observer agreement, intraclass correlation coefficient (ICC) was reported. Since our cohorts were small, non-parametric statistical analysis was applied in this study. For in-vitro verification, PD from 4D flow MRI and CFD were compared using Spearman correlation analysis. The symmetry in FR and PV for each vessel and for hemispheres were examined by performing one-sample Wilcoxon rank-sum tests with null hypothesis that the ratios would be one (i.e., symmetry). The relationships of PD with FR ratio and PV ratio were examined by Spearman correlation analysis. To investigate whether the hemodynamic parameters were sensitive for differentiating moderate and severe stenoses, FR ratio, PV ratio, and PD between the two subgroups were compared using the Wilcoxon rank-sum test. All statistical analyses were performed in MATLAB and P<0.05 was considered statistically significant.

Results

Study Cohorts

A PACS-based search for ICAD and intracranial stenosis patients who underwent dual-venc 4D flow MRI identified 22 subjects. Six patients were excluded from the study: two patients presented with total vessel occlusion, two patients due to insufficient coverage of the CoW, one patient with stenosis involving the M2 segment of the MCA (small, <2 mm vessel), and another patient due to extreme velocity aliasing in the high-venc acquisition which was uncorrectable. Therefore, a total of 16 ICAD patients (4 women, 61±14 years of age) were included in this study. The demographic information of patients and volunteers is summarized in Table 2.

Table 2.

Demographic information of healthy volunteer and patient cohorts

	Healthy control	ICAD patient	p-value
No	25	16
Age (yr)	41 ± 18	61 ± 14	<0.001
Gender (female/male)	10/15	4/12	0.08
Height (cm)	173.8 ± 14.1	173.0 ± 13.0	0.97
Weight (kg)	81.6 ± 20.0	85.9 ± 22.2	0.62
Body mass index	26.8 ± 4.5	28.4 ± 4.8	0.36
Hypertension	-	14 (87.5%)
Diabetes	-	10 (62.5%)
High cholesterol	-	11 (68.7%)
Symptomatic	-	15 (93.7%)
Stenosis location
ICA	-	3 (18.75%)
MCA	-	13 (81.25%)
Degree of stenosis
Mild (ICA/MCA)	-	2 (0/2)
Moderate (ICA/MCA)	-	7 (1/6)
Severe (ICA/MCA)	-	7 (2/5)

In-vitro Verification of Pressure Drop Calculation

As shown in Figure 3a, 4D flow MRI overestimated the PD by 0.6–3.0 mmHg compared to CFD results. However, the increase of PD with FR and the degree of stenosis between CFD and 4D flow MRI followed similar trends (Figure 3a). The relation of PD obtained from 4D flow MRI and CFD simulation is shown in Figure 3b, demonstrating a strong correlation between the two results (ρ=0.98, P<0.001).

Figure 3.

Comparison of 4D flow MRI vs CFD for the pressure drop calculation in stenosis phantom

Comparison of Manual and Semi-automated Flow Analysis

PV and FR in arterial branches of the CoW were determined both manually and using the semi-automated tool by two observers. In total, 45 vessels were evaluated (5 subjects and 9 vessels each). For manual and semi-automated comparison, the ICC of PV and FR were 0.49 and 0.925, respectively. The ICC for inter-observer agreement for the pair of (PV, FR) was (0.38, 0.92) and (0.83, 0.99) for manual and semi-automated analyses, respectively.

Cerebral Blood Flow Distribution in Controls

Mean ± standard-deviation of area, FR, and PV across the cohort are presented in Figure 4, demonstrating the normal ranges in the arteries of the CoW. Figure 5a depicts the vessel PV ratio and vessel FR ratio for controls, demonstarting there was no difference of PV and FR between right and left arterial branches with vessel PV ratios and vessel FR ratios approximately equal to one (Table 3). These results suggest symmetry of FR and PV in major cerebral arteries of the brain in controls. In addition, FR and PV between two hemispheres were the same (Figure 5c and Table 3).

Figure 4.

Schematic diagram of the CoW demonstrating cumulative results for healthy cohort (mean ± standard-deviation of the entire cohort)

Figure 5.

PV and FR ratios (a and b) per individual artery and (c and d) per brain hemisphere in control (left panels) and all patients (right panels)

Table 3.

Summary of the symmetry ratios in controls and patients.

		vessel symmetry ratio								hemisphere symmetry ratio
		ICA		MCA		ACA		PCA
		median	IQR	median	IQR	median	IQR	Median	IQR	median	IQR
Controls (n=25)	PV	0.97	0.18	0.99	0.17	1.05	0.44	1.06	0.19	0.97	0.15
	FR	1.01	0.19	1.01	0.13	1.10	0.59	1.02	0.17	1.02	0.20
Patients (n=16)	PV	1.05	0.41	1.31*	0.73	1.08	0.34	1.10	0.35	1.32*	0.30
	FR	1.04	0.20	0.76*	0.30	1.01	0.41	1.03	0.27	1.01	0.35

^*significantly different from one

Hemodynamic Impact of ICAD in Patients

The vessel PV ratios and vessel FR ratios as well as the hemisphere PV ratios and hemisphere FR ratios in patients are summarized in Table 3. It was found that vessel PV ratio and vessel FR ratio were significantly different from one in MCA (P=0.04, Table 3), while the vessel ratios were approximately one elsewhere (Figure 5b and Table 3). The hemisphere PV ratio and hemisphere FR ratio are presented for ICAD patients in Figure 5d. The hemisphere FR ratio was symmetric for ICAD patients, but the hemisphere PV ratio was significantly greater than one (P=0.002, Table 3).

As shown in Figure 6a, there was a significant correlation between stenosis PD and hemisphere PV ratio in patients (ρ=0.67 and P=0.02). In two severe cases, the stenosis was below the resolution of MRI and could not be imaged. Therefore, although the data for those subjects was still suitable for the analysis of blood flow redistribution in the CoW, PD was estimated for the remaining five severe stenosis patients.

Figure 6.

(a) Correlation between PD at the stenosis and hemisphere PV symmetry ratio. Mild, moderate and severe cases are shown by the black squares, blue circles and red triangles, respectively (b) Hemisphere PV and FR ratio and (c) pressure drop for ICAD patients with moderate and severe stenosis.

The results for hemisphere PV (and FR) ratios and stenosis PD for two patient subgroups of moderate and severe stenosis are shown in Figure 6b and 6c and summarized in Table 4. It was found that while PD was significantly (62%) higher in the severe group (P=0.02), hemisphere PV (and FR) ratios were similar (Table 4). The mild subgroup included only two subjects and therefore was too small for subgroup analysis. However, as shown in Figure 6a, stenosis PD increases from mild to moderate and severe stenosis subgroups.

Table 4.

Summary of the symmetry ratios and PD in patients with moderate and severe stenosis.

	Moderate (n=7)		Severe (n=5)#
	median	IQR	median	IQR
Hemisphere PV ratio	1.31	0.34	1.45	0.41
Hemisphere FR ratio	1.09	0.31	0.95	0.17
Pressure drop	2.3	1.30	3.73*	1.09

^*significant difference between subgroups

^#two severe subjects were not suitable for PD estimation due to signal loss at the region of maximal stenosis

Additionally, it was observed while in majority (88%) of healthy controls the maximum PV located within ICAs, the maximum PV in 100% of severe, 83% of moderate, and 50% of mild MCA ICAD patients was detected at the location of stenosis in MCA.

Discussion

A semi-automated analysis tool is presented, which enables multi-modality MRI (TOF and 4D flow) post-processing for user-independent lumen segmentation and flow quantification simultaneously in all major cerebral arteries. The presented tool is an extension of centerline processing scheme, which was shown to be effective for the assessment of blood flow in intracranial arteries (50). The findings of this 4D flow MRI study show that interhemispheric PV ratio and stenosis PD could be useful for hemodynamic assessment of ICAD.

Alternations in cerebral hemodynamics due to ICAD have been investigated previously using digital subtraction angiography (DSA) or with positioning of multiple 2D PC MRI planes across the CoW (7,15). DSA is qualitative and invasive, and the injection can affect blood flow (58). Moreover, risks of catheterization, radiation exposure, and neurological complication (59) make DSA less favorable in the absence of endovascular intervention. 2D PC MRI is limited with insufficient anatomic coverage and the manual placement of planes limits the reproducibility of results and increases overall acquisition time. ICAD is a dynamic disease with a high rate of stroke recurrence in symptomatic patients, demanding regular follow-up with non-invasive and reproducible methods. 4D flow MRI provides non-invasive velocity measurement across the cerebral vasculature from a single 10–15 min volumetric scan with no overhead time for multiple 2D plane placements.

Dual-venc approach enables 4D flow MRI with a large velocity dynamic range (up to high-venc for each velocity component) and high velocity-to-noise ratio (41,42). Two sets of phase images were acquired. The high-venc data is to correct aliasing in the low-venc data, so it is desired to ensure no aliasing occurs in high-venc data. In this retrospective study, we used a preset high-venc of 120 cm/s because 4D flow MRI was part of a clinical protocol. From our experience with previous 4D flow MRI studies in patients with ICAD (16), we learned that maximum velocity in the stenosis would hardly exceed 120 cm/s. This setting was suitable for majority of cases with one exception that was excluded. Furthermore, using a low-venc which is half of the high-venc would result in a good trade-off between velocity dynamic range, velocity-to-noise ratio, and anti-aliasing performance. In this study, 4D flow MRI was acquired post-contrast (~30 minutes) as it was part of a clinical protocol that included administration of contrast agent (CA) for other clinical sequences. Although 4D flow MRI is a non-contrast technique, it may benefit from enhanced signal-to-noise ratio due to CA. This in addition to dual-venc approach increases the accuracy of blood velocity measurements in smaller or stenotic vessels.

Our semi-automated flow analysis tool improves data utilization by taking advantage of volumetric information and can facilitate the transition of 4D flow MRI into clinical settings. In this study, the analysis of cerebral blood flow distribution was performed in a total of 41 subjects, which could be very tedious and user-dependent if done manually. The multi-plane-per-vessel approach implemented in the analysis tool increases the reliability of findings and prognostic value of flow-derived parameters for cerebrovascular diseases. The comparison between the manual reference and semi-automated results showed that inter-observer agreement in PV measurement was significantly improved (ICC of 0.83 versus 0.38).

A flow phantom study combined with CFD modeling was conducted to examine the Bernoulli equation for estimating PD for ranges of FR and degree of stenosis. This in-vitro experiment was not designed to replicate actual physiological flow. Similar to the approach adopted in (36,38), our purpose was to establish a simple experiment where we could rely on results from both CFD (without simplifying assumptions) and 4D flow MRI (less affected by spatial resolution). Thus, the PD algorithm could be effectively tested in a simple known geometry under well-controlled flow condition. The results of the phantom study demonstrated discrepancies (< 3 mmHg) between CFD and MRI results, which is smaller than the limit of agreement (> 8 mmHg) reported in Ha et al. (38). Since the overestimation was consistent and there was a strong correlation between the two methods, results for PD can still be compared among cases, and relative variations of PD may allow assessment of stenosis severity. In this semi-automated analysis tool, we used a simple method for PD estimation to enable clinically useful analysis of 4D flow MRI.

The flow in the phantom was turbulent (56), so we performed CFD simulations with turbulence modeling to accurately simulate flow in phantoms. 4D flow MRI sequence could be also modified for turbulence assessment by adding six extra non-orthogonal velocity encoding (60). However, the phantom study was to examine the same 4D flow MRI sequence as applied for clinical scans. Our study was a retrospective analysis of 4D flow MRI data from a clinical scan where the scan time was an important concern. It was shown that PD calculated with the Bernoulli equation agrees well with PD obtained from turbulence-based PD estimation methods (36–38).

Analysis of flow redistribution in the CoW may indicate collateral flow pathways (61). It was suggested that direct (CoW) or indirect (retrograde pial) collateral flow can alter outcome and risk of stroke in ICAD (62). Ruland et al. (63) reported possible PCA-to-MCA collateral flow pathway in stenotic MCA patients. Wu et al. (16) also proposed PCA compensation in addition to increased blood flow in the ipsilateral ACA in patients with MCA stenosis, which was in agreement with findings of Brass et al. (64). We observed similar changes in a number of ICAD patients. However, the comparison of FR and PV between ipsilateral and contralateral vessels may be unreliable due to anatomic variations in the CoW (53). For instance, there was an outlier in the moderate subgroup (Figure 6a, PD of 3.79 mmHg and hemisphere PV ratio of 2.56), where the blood flow redistibution may be influenced by two opposing phenomena. The subject presented with an aplastic A1 segment of the contralateral ACA, a normal CoW anatomic variant, resulting in increased blood flow to ipsilateral ICA and ACA (53), as opposed to the stenotic MCA effect that is associated with blood flow reduction in ipsilateral ICA (16). Therefore, in this patient, the FR and PV of ipsilateral ICA increased instead of decreasing. Furthermore, it is unclear whether the flow increase to ipsilateral ACA was due to collateral flow or aplastic A1 segment of contralateral ACA. This example demonstrates that asymmetry in blood flow distribution may not be necessarily related to underlying vascular disease and indicates pitfalls of interpreting flow findings in the absence of cerebrovascular anatomy knowledge.

Hemispheric flow comparison may provide more reliable assesment of hemodynamic status (22). Our findings showed that interhemispheric FR ratio in ICAD patients was approximately one (i.e., symmetrical), which could be due to collateral blood flow via other ipsilateral cerebral arteries to compensate compromised flow in the affected vascular territory (10).

Several studies have investigated new criteria based on blood velocity and flow in the stenotic vessel to assess stroke risk (13,14,20). Amin-Hanjani et al. (15) applied QMRA flow assessment in vertebrobasilar ICAD and demonstrated increased stroke risk associated with low flow status. We investigated stenosis PD which was previously shown to be useful for stroke risk stratification (18). We demonstrated that PD elevates with diameter constriction from mild to severe stenosis. However, unlike the degree of stenosis that is static and only sensitive to diameter change, stenosis PD also increases with blood velocity. If the blood flow demand through stenotic artery is high, the PD would be high meaning the stenosis would be associated with a high risk of causing flow compromise in distal vasculature. In this feasibility study, good agreement was found between classifications based on the degree of stenosis and PD. However, future studies with larger cohorts in each subgroup are warranted to investigate the effectiveness of PD for assessment of ICAD.

This study associated with several limitations one of which is small size of patient cohort. This is a feasibility study to establish imaging and analysis methods, so we combined subjects with ICA and MCA stenoses in one cohort, which may introduce some heterogeneity in the patient cohort. To minimize negative impact, hemodynamic asymmetry in ICAD patients was evaluated using flow parameters defined per entire hemisphere (i.e., hemisphere ratios). Additionally, patients and controls are not matched for age and gender. The absolute values of FR and PV are age-and gender-dependent (65), so the FR and PV ratios served as normalization to allow comparison between the patients and controls.

Another limitation is insufficient spatial resolution of current 4D flow MRI for small intracranial vessels. In a phantom study, we found that applying dual-venc 4D flow MRI with three voxels across diameter may result in 10–15% inaccuracy in flow (66). However, at the site of stenosis with diameter smaller than three voxels, the blood velocity measurement may be influenced by partial volume effects. To the best of our knowledge, currently there is no other technique capable of capturing the velocity field across the entire cerebrovascular network that enables velocity measurement with higher accuracy. For instance, typical slice thickness of 2D PC MRI (6 mm) would result in even lower accuracy than 4D flow MRI. In this study we only included patents with stenosis in larger arteries and excluded those with stenosis non-resolved on MR images. The significant difference in PD between moderate and severe ICAD subgroups suggests that despite this important limitation, the method is still sensitive for assessing the lesion.

Our flow phantom study was also limited due to steady flow condition. In this study, we used CFD as a reference to examine if the Bernoulli equation and 4D flow MRI can provide good estimation of stenosis PD. Therefore, a simple and well-controlled flow condition was considered to achieve reliable results from CFD. Future phantom studies with pulsatile flow and comparison with pressure sensors are warranted for validation of our methods under more realistic condition.

Conclusions

This study demonstrated the feasibility of dual-venc 4D flow MRI and a semi-automated flow analysis tool for evaluation of intracranial hemodynamics in patients with ICAD. The hemodynamic significance of ICAD was characterized by cerebral blood flow redistribution and PD at stenosis. As a baseline, the flow analysis in healthy volunteers showed symmetry of FR and PV between the brain hemispheres. Then, it was found that in patients, PV on the affected hemisphere was significantly higher than the normal side, and FR in both brain hemispheres was the same. It was also demonstrated that PD at the stenosis was significantly higher in severe stenosis compared to moderate stenosis patients. The efficacy of PD in differentiating severe and moderate stenosis and its correlation with hemisphere PV ratio suggest that PD could be useful for non-invasive assessment of the hemodynamic significance of ICAD.

Supplementary Material

Supp figS1

Supporting information Figure S1. Variation of peak velocity along the centerline of a stenotic ICA before and after applying the median filter. The region of the stenosis is shown